Healthcare Data



hello world it's Suraj and health care is the most exciting application of AI technology using AI we can cure all major diseases help doctors treat more patients and even enhance our own cognitive abilities to superhuman levels the school of AI just announced its first global hackathon called health hack at a launch event in Singapore's national gallery so I flew to Singapore to give a keynote on AI in healthcare at the event as well as attend a panel discussion on the same topic in this video you'll get to see exclusive footage of that keynote and the panel discussion I hope you find it educational and inspirational and if you're new subscribe to get notified when I release my videos enjoy introduce you to suraj revolt the founder of the school of AI as I said an international nonprofit with the mission to offer a world-class education in AI to anyone on earth for free Suraj's YouTube channel of AI education content has over 500,000 subscribers and he partnered with Udacity to launch their deep learning foundations nanodegree program besides being an AI educator Suraj is also a best-selling author public speaker data scientist contributor to open source projects like open AI and rapper so please join us in welcoming Suraj of all thank you everybody thank you thank you let's see how loud this mic is okay perfect good very good okay I'm very excited to be here in Singapore yesterday I flew in from Los Angeles it was a 17-hour flight so I'm here for a reason I believe in the people here I think we can do a lot of great things this community here in Singapore can do a lot of great things for health care and so in this talk I'm going to talk about AI in healthcare specifically I'm going to talk about AI in drug discovery and a few other applications I'm gonna list three applications my thoughts on them and then we're gonna talk about for each of these applications how this would work using technology we're gonna get into specifics okay so before I start I want to talk about what what I'm most excited about what I'm most excited about is the convergence of biology and information so if we think about it the gabite the gene and the atom let's just think about that for a second the byte the gene and the atom these are three fundamental building blocks for three different subjects the byte for digital information right the bytes from the bytes comes kilobytes megabytes and then programs and algorithms and everything else data from the gene you know we all have gin we all have DNA genetics a lot of it help makes us who we are our traits over into how how we think a lot of that comes from genetics and so biology from from the from DNA from genetics Springs life and the third is the atom so matter right a lot of amazing innovations have come from matter and so in the 20th century we saw a lot of incredible innovation segmented in each of these fields so in biology we saw revolutions in health care in terms of cures for new diseases and ways of dealing with genes such that we can create new gene therapies different things like that when it came to the bite I mean I don't even need to explain all that's happened in computer science in the past five decades and then for the atom you know physics geophysics I mean earthquake prediction there's so much there as well so that's what defined progress in the 20th century these three fundamental building blocks and in specifically how we humans use those to make innovations but I'm here to say that in the 21st century where we are right now what's going to define progress is the convergence of these three fundamental building blocks so the one convergence in particular I'm very excited about is the convergence of the bite in the gene of information and biology because we have because of computer science because of data because of algorithms we have so much biological data that is now available to us you know I just had a test done and I had blood drawn and it was like these three vials and the lady was taking away I was I hold on I want that data she didn't even understand what I was talking about these these three fields are segmented there they're separate but they need to converge because blood my blood has so much data I want to know about myself and and and and what I'm into what what how I can optimize my life better using what's built into me like what types of food I should eat what types of activities I should be doing what types of climates I'm best suited for things like that and algorithms information can help approximate those solutions but that's just one example we have several coming up so that's I just wanted to give you a kind of base frame of thought as we go into this AI in healthcare space we got to be thinking about the convergence of the bite and the gene in particular that's what that's what AI and healthcare really is so first application is drug discovery so so specifically in the u.s. it takes about ten years for a drug to come to market and it takes tens of millions of US dollars and so that is I mean it's generally long wherever you are in the world because of different regulations and that just that's just how it is that's just how it has to be it's not like some you know nefarious plot that's just how it is because of the process of having to select the best candidates for you know some treatment for some disease you know a lot of molecular biologists and and people working in cancer therapy they have to pick from a you know 11 million plus candidates for what could be the ideal cure for whatever disease are trying to tackle and roughly they're doing this by hand but AI can look at this information and scale that discovery in a way that we humans can't write algorithms digital digital algorithms digital intelligence can look at data in a way that we can't and what this can do is it can speed up this process this pipeline from basic research to clinical trials all three phases to the review process and until it finally becomes a medicine that can save people's lives so let's talk about specifically how this would work you know technically speaking right listen let's not just talk about buzzwords how would this work so let's say we have a we have a genetic database for gene therapy we would be using a genetic database in the case of let's say cancer which i think is the most exciting because this is a major unsolved problem we have a list of a data set of possible drugs that could be used to treat you know whatever cancer you're trying to treat and the idea is that how do we find the needle in the haystack in that how do we find the one that sticks out from the rest that's going to be perfect for this specific disease that we are targeting so that would be a discriminatory problem we are trying to discriminate amongst a bunch of potential drugs to find which is the one that could cure the disease but what would be more interesting is framing it as a discriminatory problem we could frame it as a generative problem what we're trying to generate the cure by using the existing data set and applying some sort of distribution to it so what I mean by distribution is we take that initial data set and we think of that as a bunch of ones and zeros so in machine learning we think of this input as a matrix and what a matrix is is it's just a group of numbers it's a giant group of numbers it can get big you can get as big as you want and we take this group of numbers and then we apply a series of multiplication operations to it that's that's what AI is it's a series of multiplication operations until the initial input which is a group of ones and zeros becomes an output which is a different groups as ones and zeroes and from those ones and zeros this translates to a drug right so think about it the bytes in the gene DNA blood the basic building blocks of life this is information so we are using information that we create with electronic equipment to find the information for cures to diseases and so if we frame it as a generative problem we can generate new types of drugs that would cure potential diseases in particular a model an algorithm that I find to be very promising in this field or they're called generative adversarial networks general adversarial networks and the idea behind these networks is that we have two competing models one is trying to generate new candidates for a drug in the case of drug discovery the other one is acting as a discriminator and it's looking at this what it outputs the you know the the group of ones and zeros and it's saying is this secure or not is this is this valid or not and so this these two models play a game well one's trying to generate a cure and one is trying to discriminate whether it's valid or not and that's a very high-level idea behind these two mathematical models by the way all of this is math all of this is it's just math it's matrix math that's it it's nothing it's it's not more than that it's just Matz and so people need to understand that when we're talking about this is a bit of a tangent but it's a necessary tangent if we're talking about AI and apocalyptic AI and all this stuff we're talking about the potential for humans to use math specifically matrix math to either exploit other humans or to help other humans empower other humans that's that's the the issue is humans it's not matrix math it's not AI AI is just matrix math okay so that's a tangent back to this so generative adversarial networks they play this game generates discriminate and I think this is a very useful model to use in the field and so in this hackathon I'd like to see people try this out right so this is readily available I have several videos on this there's a bunch of free tutorials on the internet about this so if you search GA and tutorial healthcare a bunch of great tutorials are gonna come up so that's the first application that I'm very excited for because there's so many people who have been affected by this you know hi in particular I have someone who I care about who has been affected by cancer and so it's it's a both a personal thing and a general thing as well so that's a very exciting one the second one automated diagnosis so if we think about the process that a doctor uses to diagnose a disease they are using their five senses they're using their human vision biological vision they're using their sense of touch right they're using their sometimes even smell different things like that and so to try to diagnose what a disease is and again this is a discriminatory problem they have a set in their head and there are biological intelligence they have a set of potential diseases that this could be and they're trying to discriminate based on the input data right what what diseases could be and so if we frame it like this that the reason I'm repeating this is because we have to frame it as all of this is just information healthcare it's all information digital genetic information it's all information and so if we think about that process how do we automate this electronically and how do we help help doctors help make their lives easier this is not something that will replace doctors it could if we want it to but if we frame it as how do we aid doctors and and make it so that they can scale the number of people that can help then this could be a good thing for doctors and then this could be something that doctors would embrace so one example in particular radiologists right so radiologists there most of their job 80% of their job is to look at these images and use their biological vision to try to decipher whether or not someone has a particular disease has now been proven to be better objectively than radiologist at this what a radiologist could do with this technology is they could scale the number of people they could be treating at the same time in a lot of countries across the world there's not enough doctors to treat you know the number of people there are so this is just a scaling technology for doctors in that that's one thing that it does so automated diagnosis we're using convolutional networks convolutional networks are a type of again matrix math combination it's a particular set of operations between that input data you know all of these all AI is just a different set of operations between for input data sometimes you do three multiplications and then you do a division sometimes you do two multiplications and then you do you multiply it by a specific type of function right so it's just it's just mixing up these different operations to see what would be the ideal for whatever our goal is and in this case our goal would be automated diagnosis so the input data in this example here are a series of brain scans for Alzheimer's and so radiologists and you know neuroscientists and you know different types of brain doctors are looking at this manually with their with their biological vision but again this is information this is information we can do this better and so automatic diagnosis is something that I'm you know very excited for in particular and we can use again convolutional networks which are a type of matrix math operation set that's useful for classifying images and saying what is in this image so if we have a data set of Alzheimer's images with a label so you know think of it as an Excel spreadsheet one column is just a bunch of images of a brain scans and the other column is just yes or no is that does this patient have Alzheimer's or no right so just two columns and that using that data set we give it to a convolutional neural network by give it I mean convert that into a series of ones and zeros multiply it by whatever model were using to use in this case convolutional Network it's going to give an output and the learning process here what it's doing while it's a learning these mathematical models they have a set of again matrices of ones and zeros that through this optimization strategy it's updating itself by updating by learning what is really doing is it's strengthening the connections between different parts of this input image to know what it means to have Alzheimer's and what I mean by saying what it means is what parts of that image contribute to Alzheimer's what what what is the essence of what Alzheimer's is and it represents this as a series of ones and zeros it learns this representation and so we we keep a representation in our head after you know if we have gone if we're a doctor and we've gone to metal medical school by looking at many many many images of this and so we can replicate this electronically as well so automatic diagnosis is a very exciting application of AI in healthcare and convolutional networks are just one type of model many other types of model and this is by far this is not a solved problem at all so there's so much all of these above by the way we like we're literally on the precipice of this new revolution of the intersection of biology and information this is a really exciting time to be in this space right now so that raised me to application 3 which is personalized medicine I mean like I said when I got my blood test done and I was just looking at that I was thinking like this is data this is data all of this is data and I want this data because I want to help optimize my life I want to get into I want to not have to go to the hospital I want to have this preventative assistant telling me what I should and shouldn't be doing that has my best interest in mind you know if maybe you know I'm about to eat something and it's like no don't do that you know you should eat this or something like that right so these are just ideas of how this could work but the idea behind personalized medicine is doctors are expensive not everybody has access to a doctor so if we can automate the whole diagnostic process into a personal assistant that everybody has on their phone and everybody has a phone and everybody will get a phone as the cost of these machines drop drastically eight million people will come online this year which is very exciting they're going to have access to personalized medicine if we can put this technology again this set of matrix operations mathematical operations onto a server that they can access via a website upload their information and then they have their own doctor in their pocket so that's super exciting as well our goal as an organization at school of a is to help solve the 17 what are called sustainable development goals which I'll talk about at the end but this is a big one this is one of the sustainable development goals outlined by the United Nations so there's one more I want to talk about and that's the most exciting to me it's also the most futuristic but that is synthetic biology so again the bite and the gene genomic data is really interesting there are so many applications to create new types of not just gene therapies when we think about genetics we think about generally we think as up as the public we think about using genetics to try to fix some problem that we have but what I think is more exciting is not gene therapy but gene enhancement what if we could use our genetics to create some new type of drug that could increase our intelligence by orders of magnitude and by intelligence I mean our ability to learn children are much better learners for language than we are as adults but we can replicate that as adults what if we could be like sponges and learn you know Mandarin in a week with this with this new type of drug so super intelligence is something that I would be really interested in new types of gene enhancement not just intelligence but endurance reversing aging these are things that are sound very science fiction but again there's so much possibility when it comes to combining biology and information another one that I find in particular very exciting you know I was just on the plane and I watched five movies because it was such a long flight one of them was Jurassic world and I was looking at the idea and I was thinking in my head I was like okay in this movie the scientist found this sample of DNA and they reconstructed a dinosaur from it but the reality is that the half-life of DNA is something like 520 years but dinosaurs were around 60 million years ago so it's all that DNA is gone now and so a lot of leaders in that space have said that oh you know we just can't do that because the DNA is gone but what they haven't thought about is using are using generative models not necessarily use that DNA to add difference splicing whatever's missing but the completely regenerate that DNA from scratch using the DNA that we already have from genomics data so think about it as reverse engineering what their DNA would be from the existing DNA that we have and so yeah we could have Jurassic Park be real I think it's possible so I hope to see a startup like that in some at some point but also synthetic biology is very interesting it's it's super-super like we haven't even begun to think about the possibilities of say molecular computing I mean that that's kind of out there but use DNA can store something like 200 terabytes I'm just kind of guesstimating here but like a lot like a single strand of DNA can store and does store terabytes and terabytes and terabytes of data and if we could use if we could somehow modify DNA to both store data and to process data we wouldn't need these machines these phones in our pocket anymore they would seem like ancient and they will in the future and by the future I mean hopefully in the next 20 years or so but using DNA as not just data but as a mode of computing and for storage is wildly exciting because think about deep learning right so in 2012 that's when the deep learning revolution started abusing these 30 year old models all of a sudden they just became amazing at image classification all these other tasks and the reason for that was because we had better compute and more data those were the two reasons right because we have better compute and more data these ancient not ancient in terms of computer science they're ancient thirty-year-old models all of a sudden worked really well so imagine if we had computing that was orders of orders of orders of magnitude better than the best GPUs and CPUs on the planet then we could use these existing models and have again a whole new revolution so something like molecular but molecular computing and molecular storage has a lot of potential for creating an entirely new paradigm of computer science and opening up an entirely new paradigm of algorithms and possibilities and all of this comes from the again the intersection of the gene and the bytes by the way if you want to learn more about this the gene the byte the atom I didn't make this up this came from a book I read called literally called the gene by siddharth Mukerji who's an oncologist too definitely read this book it's a great book by the way this is gonna be super useful to you like outside of all this what I do when I read is I read books that I read books by listening to them as audio books at 3x speed and so in the past six weeks I've read thirteen books and so what I do is I just kind of like sit there I'm listening I'm not really doing anything I'm just sitting there and I'm listening and absorbing the information and I've had and I've said this on my channel on my youtube channel but a lot of people were skeptical but even now I'm getting people who are saying Suraj you don't know how much I'm learning now like I was skeptical that I could do this like 3x is way too fast and I'm not a native English speaker and you are so I can't do this well guess what you know a lot of my subscribers primarily from India as well are able to do this now not a lot like three I got three of them like tweeted to me that they could which made me very happy so so you got it you got to work up to it it's like it's like you know you're lifting weights you know you're not gonna be I couldn't do 3x when I first started you know I can only do like 1.5 max but if you start at one and then go to 1.5 and do that for like two weeks three weeks your brain your neurons your connections are gonna they're gonna update they're gonna strengthen and I see someone shaking their head no this is real this is real I'm see there's still someone skepticism around the possibility of this but it's a real thing try it out don't take my word for it try it out yourself and you'll and you'll see but this is a great way to taking a lot of information until we get to the time where we have devices and maybe gene enhancements that help us accelerate how fast we can learn this is a great midway point like listening to things at 3x speed video as well I do that even my own video so these these four applications they're part of our mission to help solve these 17 sustainable development goals outlined by the United Nations so the idea is that these are very ambitious goals in the United Nations outlined by 2030 which is actually you know like only 11 years away it's it's pretty ambitious they want to solve all these no poverty zero hunger these are like these are very ambitious initiatives and and we have school of AI believe that with this AI technology use the massive data sets that are available to us on the Internet today that anybody anymore has access to regardless of where you're born or you know what your profession is or what your schooling is you will you aren't able to make an impact with this technology and we believe that with the combined effort of the globe put together people from across the world from different cultures different viewpoints different perspectives if we put this together we can come up with these incredible solutions using this technology and we can accelerate the progress towards these goals and our goal in particular is to make these happen by 2025 not 2030 and once we do that once we're able to solve these goals what that means is for them for the majority of the world as everybody comes online and everybody will be online and then within the next decade all of humanity mostly like 99% it's gonna be online and then we have an entirely new internet culture when this happens and everybody's basic needs are met then we can get to the stage of creativity and innovation globally and is just gonna build on itself and we're just going to accelerate and accelerate and accelerate it's a very exciting time to be alive in human history right now when we have the potential for these technologies to solve these incredibly hard problems and enhance our our way of life make improve our lives improve our relationships with people and improve just the way we act in the world and how we know about all the complexity in the world simplifying it such that we know exactly what to do when to do it were more confident we feel better about ourselves just overall we're trying to increase the well-being the health the intelligence the emotional intelligence of humans using this technology and we very much think it's possible thank you so now I'd like to introduce you to our panelists so first up we have Ayesha Khanna so Ayesha is co-founder and CEO of IDO I oh hey I sorry 8'o AI an AI solutions firm an incubator she's advised governments and corporations on AI smart cities and FinTech and she's on the board of Infocom media development authority which is the government agency behind Singapore's a smart nation vision Asia has a PhD in information systems and innovation from the London School of Economics as you're in two books and multiple articles and technology and innovation and has been named one of Southeast Asia's groundbreaking female entrepreneurs by Forbes magazine thank you for joining us today Aisha next up we have dr. Victor Tong Victor is the chief digital and information officer at the National Gallery Singapore where we find ourselves today he was the founding director of a star social and cognitive computing department and the head of biomedical informatics at the Institute for infocomm research he's also held senior roles at companies like Pfizer and sa P research victor has a PhD in biochemistry from the National University of Singapore and his work has won numerous awards including the MIT TR 35 Award and the World Economic Forum young global leader whose welcome dr. Victor Tong next up joining us is Charles who's spatha he leads Accenture's artificial intelligence practice in Southeast Asia he has over a decade of experience in AI across industries including FinTech advertising security and the military in the past six years Charles was founded to AI startups one focused on predictive maintenance for the oil and gas industry and the other on fraud detection and credit scoring for unbanked populations so please welcome Charles and last but not least we have dr. Tenten we is the CEO of the national supercomputing Center in Singapore where he manages Singapore's first petascale supercomputer he was also the founder of the bioinformatics Center at NUS the asia-pacific bioinformatics Network the first Internet service provider in Singapore he founded and oversaw the creation of the multilingual internet domain name system and Wired magazine has called him one of the people who built the in Internet so please welcome dr. Tenten wig and of course we'd like to invite Iker and Suraj again to join us up on stage and our first topic that we're gonna be talking about today is ethical AI so that might seem a bit vague to people who are new to this question so maybe one of you can can make that more concrete what are we talking about when we're talking about AI and ethics is this like Alexa and Siri getting together to rob a bank or something like what what is the problem I guess maybe a way of putting the problem is when you look at the way traditional software programs or somewhere systems or capabilities are built there is a very easy way for us to explain the decisions that are taken by that software there is a deterministic link between the input on the output right the moment we perhaps put artificial intelligence into the equation such link gets a bit more blurred and the decisions that the system are making in the context of think of classic example that comes to mind is autonomous driving so two autonomous cars are going to be in an unavoidable collision does the system have to make any decisions that affect who survives so who doesn't survive for example what are those considerations that are going to be taken into account how do those decisions get taken what are the parameters what are the outcomes what why do we take the decisions we take and how did the system reach the conclusion that's perhaps one one way of studying the topic is it whenever a differing opinion is is that is that the problem is the problem of inscrutable ai ai models that do things we maybe can't take responsibility and that's definitely one of the problems I totally agree I think that one can kind of divide it into three problems one is explained ability which is even for AI engineers sometimes it's hard to understand why it actually came up with it then it's interpretation which is for the business user um or for the domain expert who's partnering with the IEEE engineers and the third is ethics so are you comfortable with what it's doing and in ethics itself there is two ways one is I think when you you don't create algorithms that are deliberative deliberately manipulative such as all the deep fake algorithms and the potential over there that you keep your data processes lineage and transparency in order those are some of the recommendations by IEEE MDAs framework for ethical AI and and then the third is trying to really you know establish global ethics and then I guess cultural ethic seven different countries may have their own limitations may put all their own limitations while others may not so if you look at GDP are I think it's a I personally think it's a really good standard for the rest of the world to follow because it really puts human rights as data rights or data rights as human rights right at the core of how we approach algorithms so is there room then for research in AI to teach us more about global ethics or cultural ethics or what we want those things to be as a society I guess a bit of a tough question too big but I guess a way of looking at it is that the fact that AI exists and it's something that is totally unavoidable guess I guess it puts a spotlight on those questions and it is going to force that conversation to take place right maybe in a more elevated or intense way that it has been happening so far I mean I think the dumb way to do it is to get either a bunch of lawyers as policymakers or to get all like engineers together whereas we really need to have a diverse team we need more people who are philosophers or artists or and people who come from different domains to get together this is literally something that's going to impact society and economy to such an extent that perhaps it's even rewiring our social contracts with the government and with corporations so it requires a more diversity and the people who address it totally agree with yeah that's a great point and that we need more into first a set of people other than engineers making these very important decisions that are affecting all of us all of us that use for example social networks Facebook is the easiest example to make fun of because of the amount of industrial scale manipulation and exploitation that it that it does in the name of its business and so Facebook is Facebook's newsfeed for example it's like my favorite algorithm to trash basically because Facebook's newsfeed is an example of a runaway AI it's making decisions that even its creators don't understand at this point because of the amount of data that it's using to show you what it thinks is going to maximize your attention so their business model is is ads right are ads so its revenue the revenue they get mostly are from ads and so the longer they keep your eyes on the screen the more money they make and this is bad because if we look at you know a lot of studies of children who are using social networks a lot it causes anxiety and depression this is this is a proven thing it's not just a fake thing so being on social networks too much is objectively bad for your mental health because because mainly it's optimizing for your attention whereas ideally it would not be optimizing for your attention it would be optimizing for your time well-spent what would benefit you the most what is the content that would help you with whatever your objective in life is what would be educational what would not be violent but instead promote cooperation and empathy and things like that then we're not there yet because we just built this system and we just built this billion person plus scalable AI algorithm and trained it on maximizing attention but maybe in the next phase we're going to clean up what I consider to be digital pollution and instead make it cleaner cleaner data cleaner information that is going to optimize for your time well-spent and I would encourage you all to look up literally time well spent because this is an actual movement I seek at the end of the day it's it proves also something is that many of us concern weigh existing pree AI actually if we talk about credit scoring another type of application well if you consider look putting the the location of an individual into a decision-making system it is not fair because it's not because you're living in a poor neighborhood that you're not someone trustworthy and yes this is a problem that is being raised with AI concern but does that mean that he was not existing before so maybe at the same time as we're working on those frameworks for the AI is something even on the on our humanity on are something to consider also because at the end it's kind of replicating a lot of the things even regarding Facebook feeds well Facebook feeds is only replicating what TV channels are also doing trying to optimize the revenue and in a certain way because with easy-to-eat concern content so that's I think it's a great question to ask and it's on not only concerning a is a society issue here yeah that's that's a good point about you know these being pre-existing issues are there's this now very well publicized phenomenon of judges sentencing more harshly right before lunch than right after breakfast and so you know that's the same kind of bias that you can get with with humans that you could get with machines that are trained on bad data or you know that with without those precocious in mind very carefully but here again we have also to be carful correlation versus causation because if you if you take numbers like that with this conversation with my team earlier this week you can have a statistics that tells you that you have more murder when you have more ice creams being sold so we also have to be careful on those cuz of heat and stress so so yah and answerers you were talking about that we need more people we this is more than just problem for engineers or lawyers and we need more diversity and this is something that's gonna affect everyone so maybe doctor tone you can tell us what do you think is an institution like like the National Gallery where we are here what is that the role of such institutions in helping educate the public and and and helping ensure that we see the more positive sides of those results than the more negative ones do you or do you see that that institutions like the National Gallery have a role in that well I guess varnish I carry were looking at the application of AI for to improve or optimize our process so we do not really go into the aspects but I guess a fundamental problem I talk about ethically I because about the cultural differences in which for example may be different across the different organizations or even across different segments of society so how do we actually reconcile these things because when I talk about ethics if I taste refers to different has different meanings to different populations or to different people that how do we actually not get everyone on board to agree on certain sets of principles or course of practice yeah and I suppose that that's another one of those problems that exists before AI and will exist afterwards so maybe we can maybe we can transition now to the the topic of health care and and some of the implications of AI for health care so we talked Siraj talked about the impact of eighty of AI on on radiology in particular but and that potential for for personal assistance and stuff like that what are some of the areas where we're starting to see impact of AI in in healthcare or in bioinformatics or any of those areas right now well I think the most of the thing I've been brilliantly covered by a Siraj earlier so please refer to what I've been said before but I think in general we can see that AI is helping at predicting what's gonna happen next being being able to detect slightly before doctor the first signs of any kind of cancer any kind of health issue being about optimizing this is where he dreams a lot of values assisting doctors but also assisting you in trying to predict and smoothsub you all the way you leaves in order to optimize so you get the the average person gets superpowers essentially it gets medical medical superpowers I don't know if it's a superpowers but it's a statistical knowledge about okay if I do that statistical superpowers adding onto some of the comments that you are shared earlier on on cancer being one of the key applications here I have you look at the Ansari I think a lot of us have experienced this in our families and in our relatives it's it's a very complex journey for the patient in a journey that that has a lot of different stages in which when you've gone through that journey you can clearly see a lot of aspects in which artificial intelligence can really play a role in helping early diagnostic for example certain types of cancer has proven to be the difference between survival or nonsurvival right so having that radiology is being able to detect with the help of that perhaps superpower of having those super computer eyes that are helping they take something in its earlier form that is totally opaque to the human eye that would be the difference between life and death for so many people um one of the situation's I experienced in this particular case was um when a patient goes through chemo you clearly tell that there's something different in that treatment do anything else right secondary effects collateral effects of the medication they're giving you they will give you a list these are the known things and that could or could not happen to you this is the phone number you call when you experience anything else so we can learn from it and we think about that right that that sounds like a space where artificial intelligence can start gathering all that data and we can continue to learn from that and it's it's it's it's a disease that puts us all into the unexplored territory perhaps or one but definitely we can see some application there well see was given a very good overview about the use of AI for track design so example what I can touch on is convex in design so you know good days vaccine we look at using dates or with virus assembly to trigger a new response but in new generation that seems for example scientists have been using a IP data API to identify regions of the viruses that can trigger t-cell response and so many of these are new biotech since for example HIV vaccines multiple sclerosis fact since answer / true that since foreign bodies are undergoing different stages of our clinical trials some increase for HIV enter phase 2 and phase 3 for no inference of virus vaccines so these are sort of potential applications that can a I can be used for that's in design so for truck design I just actually comment on the importance of the school organizations like the school of AI there's a lot of Education that needs to take place before the healthcare community can actually latch on to this extremely powerful technology because that's with a powerful technology that comes with great responsibility and and in and the whole spectrum of ethic ethics that has to guide us in the way we apply these incredibly powerful technologies and in particular AI which has the capability of surpassing human intelligence and also this possibility that an artificial thing that is divorced from human consciousness that is able to now predict things far better than anybody the so-called superpowers that you described and so it's in this is critically important for us at this juncture of time for us to really look at this incredibly powerful technology and put in place an educational process that will help us understand one of the things that we can apply to and what we shouldn't be doing right and these ethical issues that arise I touch to the very core of us as human beings human beings are conscious of and we are moral creatures and therefore if you are envisage an artificial environment where a machine that does not have consciousness now has super intelligence then that difference now starts to create a lot of impossible and very difficult situations that from day to day activities like deciding fever an autonomous vehicle were to crash should it crash this way or that way and you've got to make that value judgment in making a pronouncement of a cancer patient as to what sort of treatment he has to take or that's the end for you is it ethical to tell a person that I'm sorry it's incurable it's the end you have no hope that Rob's the person of all hope but the intelligence says yeah it's correct it's a fact do you want to say that to a person is it ethical for you to say that to a person so it's important now for us to understand as human beings with morality and in consciousness that we are venturing into a scenario where we are talking about artificial intelligence artificial superintelligence that probably is smarter than all of us at this stage and it keeps learning where as human beings we are all born clean slate and we need education now to bring us to the point where we are smart enough to be able to have this conversation so that brings me right back to why the school of AI has a very critical task to play particularly in the healthcare community I'm 29 years 19th of February at the School of Medicine ok so you introduced me as being an internet pioneer well I'm a professor in the biochemistry department but in the last thirty years I've been trying to introduce bioinformatics and along the way no traction so I did sideline projects like the internet like super computing like multilingual domain names and so on and I gained significant traction day but in my department in my school it's still a long way to go and let that's the medical school let alone the entire establishment of the healthcare system it takes a long time so even as we are really excited and I'm a technology guy right and really excited about technology but never underestimate how long it takes for us actually educate people change the mindset to adopt technologies in an ethical manner that's really interesting yeah on that on that topic of you know how it's been taking a long time in in the areas that you're involved with is there something specific about healthcare or the health care industry that that makes that the case where where for example as Suraj mentioned you know a lot of these these techniques are based on academic papers that have been published for decades and and the research that goes on today especially you know championed by organizations like open AI is out in the open and researches out in the open is there does that fit well with healthcare it does that kind of open open does the open source philosophy fit well with healthcare or is there some other reason why maybe things are it's taking longer as you said for that adoption to to take hold a lot of healthcare data is private for good reason because it's a very sensitive data and so people working with healthcare have to find a way to get this data finesse their way to having access to this data but what I think is a great startup idea that it's one of many but like I if I had the time I would be working on this but what would be a great startup idea is to offer a service to these hospitals and other healthcare providers that offers to anonymize their data and sell this anonymous data directly a peer to peer to trusted entities that they vetted so that would be like anonymized data as a service hadass as a service for healthcare so if you can prove to these and that that's a great kind of startup to do on the ground because you can physically go to these you know offices and you know work your magic and try to sell this idea and just build your repertoire and Trust overtime but if you could do that then all this anonymize data would become more publicly accessible to people so Surat you talked about the old vials of blood you donate it right you actually gave it away and the data associated with it and the metadata that is associated with it it's actually helped in the in a database somewhere in the hospital and the data belongs to them you don't get a copy you might get and I downloaded a CD now they say they give you a CD of all your scans right but just that's just about it until the medical information of a patient now belongs to me myself it's a long hard road because right now the medical profession keeps that information and for a good reason privacy issues for example in Singapore we have had a recent spate of loss of medical information and by the millions and the whole issue about what surprised my private medical information and what how it impacts on me myself my family members or more importantly perhaps my insurability right how would the insurance agents now assess the amount of premium I have to pay if they knew that I was suffering from this or that genetic disease or my propensity towards a particular cardiovascular event and so forth and with super intelligence you could almost predict to ascend a very high percentage of accuracy a statistical or B whether I'm gonna develop this other disease and will my insurance premium exceed the point where I become uninsurable right so these ethical issues now and possibly even a lot of the legal issues now become entangled and entwine inextricably with all these powerful technologies that we are trying to promote here yeah it's a really good point it's one thing to introduce AI to medical students at the hospitals but actually it's as important to introduce it to lawyers they need to study this as well or business students and what my company does is we work with a lot of large enterprises in Asia to introduce and build AI solutions for them and we see a lot of resistance in the beginning because of politics inertia all of the usual things and this idea of privacy that is that is not even technically sound sometimes just because they don't even understand whether something is available or not available or is truly protected or not and I think everybody is just afraid of their own jobs they're kind of afraid to take this first step so making it easier for them with some kind of repeated pilot or proof of concept as a lot of companies are trying now and the Singapore government is subsidizing I think is a good way forward and we need to get all small medium enterprises that are 70% of the backbone of this country we need to get them some basic education and then access to the cloud and access to software that they can upload their data in and begin to see some returns on investment I think that that's it's one is the education sector and the other is the business sector really seeing some return on investment the problem is they all expect something they saw in the movies and there's so many times when I when my team were right you know we'll write some some predictive algorithm and it was the data wasn't good it was dirty blah blah blah and then when I thank God we got like 70% it's amazing and then they work ah 70% not 99% and it's like they're obsessed with this hundred percent 99 percent number because they've been going to too many movies are looking at Wired magazine and I think that they they need to know it's a process and all of those things are really important to keep in mind when you're trying to introduce it obviously I'm very interested in seeing how to reduce it is as well as to students yeah I actually I'm really interested in that in that obsession with a hundred percent thing because that's something I've come across as well maybe this is a bit of it of a distraction but let's see if if it strikes a chord so I'm personally interested in self-driving cars and whenever I talk to someone about self-driving cars they say well how can you make the self-driving car perfect and I'm saying well I'm not perfect and I can drive a car like why are we not asking that a question about like how can I make sure that you're perf driving a car before I let you drive the car right so why is it that our expectations of the software that we build is perfection whereas their expectations of ourselves is like as long as we're math it's okay is very related to that the hype curve we see whenever a new technology is introduced right there's always like this typical curve of you know there's a lot of high becomes a buzz board everyone wants it everyone expects the world a lot of it then this is the solution meant when you realize that 100% is not gonna happen but you may get 70 and eventually we get to the point that you know what 82 is pretty good right and I think if you look at all the applications we're seeing I real like we just sell me introduction and the fear or sort of the apprehension I think it's a journey and improperly parallel to the journey for the cloud as well as an example I probably am more an easier one to to to understand a lot of companies in different degrees in every industry are moving to the cloud and I'm moving to the cloud as a journey in which say as they understand the risks and the advantages they feel more and more comfortable moving more complex more critical workloads to the cloud nothing artificial intelligence is just a journey that we see happening a very similar way and we have a lot of projects in which we are helping implement the famous chat bone the famous in customer service in capability which is very low risk nothing really did critical depends on it you can just make someone a bit happier you remove friction from a process moving into using computer vision to automate a lot of our activities but in a inherently lower risk tasks right but we're starting to see some traction in places where people are getting more more comfortable exploring putting critical parts of their business in the hands of of artificial intelligence so I guess it's a journey as you introduce so people just have to get comfortable with it yeah like a pair of shoes the artificial intelligence so so so Victor mentioned about a computational design of vaccines for example so recently you heard that our zero tolerance for any drug vaccine that we take I requires 100% you are asking about 100% why is this it's important to us if I have a child and I have to decide whether it's gonna take this vaccine or not so that has resulted say in the US where a lot of people are against these vaccinations because they are afraid that it could be leading to all kinds of ailments like including autism and so on back by arguably fairly tenuous facts but if there is a risk my child might get I might say no but result is recent outbreak of measles for example so how do you balance that so this is where all these serious ethical considerations and and legal issues as well and and whether the government has put in a law that says no you know for the benefit of the entire community we are happy to have 1% of people suffering the negative potential negative effects and that's computes to quite a large number if you have a million people and it's 1% attack a lot of people you know I had a lot of children that would be affected by it so you gotta wait it against the public health issues versus the the lack of a hundred percent Vica see of anything so these are things that I doubt if any AI system can eventually grapple with maybe AI in F be the ultimate AI ethical system but that will be another story for another day but so it is critically important for us when we talk about AI in health care that we take this into consideration and and really think through the whole process of how we go about introducing AI in a measured calibrated and nuanced manner in health care do you have an idea for applying AI to health care let me know in the comment section and please subscribe for more educational videos for now I've got to find a float tank so thanks for watching

36 comments

  1. the only reason I want to leave a comment is that Youtube AI to recomand more of this topic…BTW I love your drive and you“re an inspiration !!! Cheers From a guy in Romania !

  2. Hi Siraj, you have mentioned about creating new animal which don't have DNA and using GAN. Is this against nature and it balance. Your thoughts please.

  3. It would be nice of you if you can discuss about the 2 mil users from the Latin American Institute of Health and the deal they struck with UND's mainchain. Nice vid!

  4. …And Rapper huahuahuaha.

    Siraj #1 to teach AI thank you so much to inspire me follow this area

  5. Hi, what should be the python code to set multiple words with multiple fg colors and font style with different values in a column in a database table. Thanks

  6. Hey! Just saw you in an Accenture video about a Health Hackathon. I thought I recognized you for the thumbnail and then I saw the video, and it was you. Awesome! Good to hear about this new association with my organization 😊

  7. How many people watched the last 30 minutes in 10 minutes when he said to speed up the audio? πŸ˜‚

  8. Thank you so much for this content! It changed my life and turned me into completely different person. I tell about what I do in terms of artificial intelligence, cyberspace, software development, drones and lifestyle on my channel. I would highly appreciate your feedback!

  9. Anybody in the mood for joint preparation (and maybe teaming up) or the hackathon in Paris? Would love to exchange ideas and rocknroll together! Thought it would be beneficial to study the relevant topics (Deep Learning in medicine, medical background, etc.) a bit in advance and go through some competition simulation. Still new in these fields and think its best and more fun to learn together (peer teaching, discussions, etc.)! Anybody interested?

  10. Byte==bite==chew==turn.
    Gene==get n e == ge in spirit == life.
    Atom == Adam == ad am == toward day.
    Converge == together turn
    Cut, turn, union, flow
    Conclusion:
    They are trying to bring consciousness to ai by integrating man with ai.

  11. This guy is not human or is a meta.human… He knows to much…
    Possibly a Jesuit or Freemason..

  12. Gene therapy? Really man? Isn't data science supposed to be based on facts? Check epigenetics. Check Gene therapy in cancer treatment. Gene therapy is BS

  13. Hi Syraj, could you help me find Charles Cruspera(I know it's misspelled that's why I can't find him)? I'm interested in using AI in the oil field.

  14. 5:02 "what types of food I should eat, what types of activity I should be doing.." The important thing is not the byte, the gene, the atom, but what is this Should that you are invoking so glibly? Totalitarianism is also based on this Should, as is all human thought, if you care to examine it. Don't forget, the most important underlying issues for humanity are not matters of fun, like playing with computers. To be fair, you are aware of this distinction. "Information" is a difficult value proposition.

  15. With all the bugs and issues that come with technology, you'd have to be stupid to get a brain machine interface. That would be just begging to have your brain messed up worse than doing drugs!

  16. who wishes that video being edited by Suraj and make it clear as usual? the screen is really far πŸ™

  17. I am Aloy Aditya Sen from Radii Corporation … I am the founder of an AI startup in Kolkata, India. I would like to get added to the list of deans in my city . Would it be possible to get some AI / ML expect s or students or enthusiasts on a meetup group within my city …. Would be helpful for us at Radii Educational Commute

  18. Dr Victong Tong is the only one who makes sense. AI in healthcare can be used to assist doctors but can't replace doctors – It's just like I personally won't fly in a plane that is not captained by a human no matter how good is the auto-pilot. Healthcare in US is a defensive industry- gooduck in introducing AI in the clinical areas especially related to patient care without the help of a doctor or nurse

  19. Yo, if you want to learn more about AI with chemical structures, check out RDKit. Also, pubchem has a lot of structural, assay, and disease indication data available.

  20. 3x speed?! Damn – I like to enjoy what (podcast/book) I am listening to and I feel anything over 1.5x makes it too much like 'work' where you are too focused on missing a beat rather than just taking in the information. Maybe just got to get used to higher speeds. πŸ˜€

  21. A simple app that uses A.I to predict how many hours you need to sleep every night based on your daily activity and how much you slept in the past days.

  22. I think its time when we should use ai for environment too. This needs a lot of research like one similar to google's energy saving algorithm that they are using in there data centers.

  23. Hey brother, I need some help. I am going to do a project on food(vegetables,fruits,etc) recognition system using AI. I have to complete my project in 3 months. I know only basics of python. What should i learn? What path(or steps) should i follow? What are the frameworks and algorithms should I use? Please help me.

  24. Great presentation. I wish it were longer because healthcare is the most exciting application of AI (as you said) and there was a lot more to be said during the panel.

    My feedback is that there is no way to anonymize health data without making it less valuable. You can remove the name (which even that is debatable – sometimes names have strange psychological impacts which can lead to actual, physical effects 😊), but if you remove the location you already loose an important clue. What if there is something about a specific place that makes people sick (or, makes their immune system stronger)?

    I don't want to make this comment too long so, I'll go straight to my idea: right now we're fighting an uphill battle with hospitals and medical organizations which don't want to share data, and regulators who want protection of personal data. What if, instead, we went straight to regular people like you and I and asked them for their data? People have to understand that anonymity can't be offered but, the upside, is that we can learn so much about health and making our lives better. As a parent, I would give my data to such a project. I want the best for my children. And yes, I want a healthy, long life for myself too. 😊

    I'm signing up for the hackathon. And I'll start working on healthcare projects. Keep up the great work. As always, you're inspiring.

  25. It's the tools to put the ai into that's nearly more interesting. Artificial white blood cell robots, Artificial organs. Synthetic cells. Body cleaning micro robots. Body structural repair micro machines. How do you get the bio data and edit if, some sort of micro surgery or even micro robots that can stay connected as a swarm in a body managed by ai , Could ai even support the liver regulating our blood or adding chemicals to our blood to augment intelligent and health, even use ai to manage hormones to extend life. How can ai get a real time picture of body data to learn from or work with ? New types of micro sensors ? Synthetic organ management, can ai replace the need for bone marrow or repair bones and the body with gene editting. Could ai even repair damage to restore youth or how about synthetic skin managed by ai , What if ai slowly replaces real cells with synthetic cells over time. I think we need brain ai interface to increase our learning speed to keep up and then we can solve the other problems when we have more intelligence as a collective species.

Leave a Reply

(*) Required, Your email will not be published